Dobit to Bside: Grounding LLM Characters into Executable Gameplay
Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment · 08.11.2026 · 22 (1) · 445-449
DOI: 10.1609/aiide.v22i1.43602 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, online erschienen am 08.11.2026. Verfasst von Yunjia Guo und 1 weiteren Autor:innen.
Abstract (Originalsprache)
We present a longitudinal case study of three generations of LLM-native social games that we built and operated from 2023 to 2026: Dobit (a mobile multiplayer life simulation, closed beta), Bside: Desktop Mate (a PC desktop pet with a multi-agent runtime, on Steam), and Bside (a live mobile companion game). We analyze this lineage along two axes: where LLM character output terminates—in presentation, language, or persistent game state—and whether that output is consumed by the player or read back by the game system. Dobit closed the loop in language: open-ended routines and daily letters supported a sense of companionship but did not feed back into game mechanics. Bside: Desktop Mate moved the loop into persistent game state through bounded, executable behavior bundles connecting agent decisions to animation, navigation, dialogue, and relationship-state updates that the game's mechanics read back. Bside mobile instead organizes short, consumable AI interactions around one persistent companion, a platform-native design rather than a port of the desktop runtime. Across these deployments, we argue that structured output is not grounding unless game mechanics can read it back, that language-grounded and state-grounded loops serve different design goals, and that AI interaction loops should be designed for their platform. The case study contributes a two-axis framework for designing LLM characters as game systems rather than conversational features.
Zitieren
APA
Guo, Y., & Zhu, J. (2026). Dobit to Bside: Grounding LLM Characters into Executable Gameplay. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 22(1), 445-449. https://doi.org/10.1609/aiide.v22i1.43602
Vancouver
Guo Y, Zhu J. Dobit to Bside: Grounding LLM Characters into Executable Gameplay. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. 2026;22(1):445-449. doi:10.1609/aiide.v22i1.43602
BibTeX
@article{Guo2026,
author = {Guo, Yunjia and Zhu, Jinghan},
title = {Dobit to Bside: Grounding LLM Characters into Executable Gameplay},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment},
year = {2026},
volume = {22},
number = {1},
pages = {445--449},
doi = {10.1609/aiide.v22i1.43602}
}
EUVIMED